Translating criteria of international forest definitions into remote sensing image analysis

Forest monitoring has received increasing attention over the past decades from various international initiatives, among them the REDD+ program which crafts an incentive-based mechanism for reducing deforestation and forest degradation rates. The success of REDD+ depends also on effective monitoring systems that can produce credible and comparable forest cover estimates. If remote sensing technologies are to be involved, methods need to be developed to implement the politically negotiated forest definitions into the technical process of image analysis. We present here a new framework for translating elements of the currently discussed forest definitions into the analysis of satellite images. The framework is based on a hierarchical classification scheme which separates the process of image classification from the application of a specific forest definition. We test this approach for two contrasting tropical regions on RapidEye images which are classified using the Random Forests algorithm. The results show that the developed framework can be operationally applied on a project level and results in standardized forest cover maps with high accuracies. Furthermore, it can serve as a research tool for analyzing consequences of political decisions regarding the forest definitions as it readily enables the user to produce forest maps and estimate forest cover for different underlying forest definitions.